Multi-Objective Sparse Sensing with Ergodic Optimization

Ananya Rao, Howie Choset

Abstract

We consider a search problem where a robot has one or more types of sensors, each suited to detecting different types of targets or target information. Often, information in the form of a distribution of possible target locations, or locations of interest, may be available to guide the search. When multiple types of information exist, then a distribution for each type of information must also exist, thereby making the search problem that uses these distributions to guide the search a multi-objective one. In this paper, we consider a multi-objective search problem when the cost to use a sensor is limited. To this end, we leverage the ergodic metric, which drives agents to spend time in regions proportional to the expected amount of information there. We define the multi-objective sparse sensing ergodic (MO-SS-E) metric in order to optimize when and where each sensor measurement should be taken while planning trajectories that balance the multiple objectives. We observe that our approach maintains coverage performance as the number of samples taken considerably degrades. Further empirical results on different multi-agent problem setups demonstrate the applicability of our approach for both homogeneous and heterogeneous multi-agent teams.

Introduction

In robotic exploration missions—such as planetary surface exploration or environmental monitoring—robots are rarely tasked with a single goal. A Mars rover, for instance, must simultaneously seek out scientifically interesting mineral deposits, maintain exposure to sunlight for solar power, and navigate away from hazardous, steep terrain. Each of these objectives is often represented by a distinct spatial distribution or "priority map."

Traditional path-planning algorithms often assume that sensors are active continuously. However, in real-world scenarios, sensing is expensive. Activating a high-resolution spectrometer or a LIDAR system consumes power, generates massive amounts of data that must be stored or transmitted, and takes time. This creates a dual challenge: how to balance multiple competing objectives while strictly limiting the number of sensor measurements taken. This paper introduces the Multi-Objective Sparse Sensing Ergodic (MO-SS-E) framework to address this optimization problem.

The MO-SS-E Framework

The core of this approach is ergodic optimization. In an ergodic search, a robot’s trajectory is planned such that the time spent in any given area is proportional to the amount of information (or interest) located there. The MO-SS-E framework extends this concept by jointly optimizing the robot’s path and its sensing schedule.

Technical Components

  1. The MO-SS-E Metric: The researchers developed a new cost function that integrates three primary factors: the alignment of sensing locations with objective maps, a penalty for excessive sampling (using $L_1$ regularization to enforce sparsity), and the overall trajectory coverage relative to a combined "master map."
  2. Sensing Decision Variables: The algorithm introduces a variable, $\lambda$, which functions as an "on/off" switch for each sensor at every time step. To make the optimization computationally feasible, this binary choice is relaxed into a continuous range (0.0 to 1.0) during the solving phase and then projected back to binary decisions to satisfy the sensing budget.
  3. Objective Weighting via TOPSIS: To avoid the need for manual weight tuning by human operators, the framework employs the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). This method identifies a "Pareto-optimal" balance between competing objectives, such as scientific gain versus risk avoidance.

Experimental Evaluation

The framework was tested using both synthetic environments and real-world geographical data.

Datasets

  • Synthetic Data: Randomized maps featuring Gaussian distributions representing areas of high interest.
  • Real-World Data (Cuprite, Nevada): The team utilized satellite and aerial data from Cuprite, Nevada, to generate three distinct objective maps:
  • Entropy Map: Representing scientific uncertainty based on ASTER/AVIRIS data.
  • Shade Map: Identifying areas with high solar illumination.
  • Slope Map: A risk-based map derived from Digital Elevation Models where high-gradient terrain is avoided.

Performance Metrics

MO-SS-E was compared against two baseline strategies: Uniform Sampling (taking measurements at fixed intervals) and a Probabilistic Heuristic. The primary metric for success was the "ergodic error," which measures how well the robot's sensing locations matched the underlying interest distributions.

Key Insights and Results

Joint Optimization and Efficiency

A critical finding of the research is that joint optimization is significantly more effective than decoupled planning. By deciding where to move and when to sense simultaneously, the robot can steer its trajectory toward high-value regions specifically at the moments it plans to take a measurement. This allows the system to maintain high coverage performance even when the sensing budget is reduced to as little as 10% of the total trajectory time.

Multi-Agent Coordination

The framework naturally scales to multi-agent teams. In tests with up to 10 agents, MO-SS-E successfully coordinated both homogeneous teams (where all robots have the same sensors) and heterogeneous teams (where different robots carry different sensors). The sparsity constraint ensures that the fleet distributes sensing tasks efficiently across the environment without redundant data collection.

Robustness to Low Budgets

On the Cuprite dataset, MO-SS-E consistently outperformed the probabilistic heuristic, particularly at the 10% and 25% sensing budget levels. This suggests that the framework is particularly valuable for resource-constrained missions where every measurement must be strategically placed.

Limitations and Future Work

While MO-SS-E represents a significant step forward in constrained robotic search, several areas for future research remain:

  • Parallel Sensing: The current model assumes a robot uses only one sensor at a time. Future iterations will explore sensor fusion, where multiple sensors can operate concurrently.
  • Dynamic Mapping: The framework currently requires objective maps to be known a priori. Integrating the system with real-time mapping and "active learning" would allow the robot to update its objectives as it discovers new information.
  • Sensor Noise: The study assumes perfect, noise-free measurements. Incorporating sensor uncertainty will be necessary for deployment in high-noise real-world environments.

Summary updated

Abstract

We consider a search problem where a robot has one or more types of sensors, each suited to detecting different types of targets or target information. Often, information in the form of a distribution of possible target locations, or locations of interest, may be available to guide the search. When multiple types of information exist, then a distribution for each type of information must also exist, thereby making the search problem that uses these distributions to guide the search a multi-objective one. In this paper, we consider a multi-objective search problem when the cost to use a sensor is limited. To this end, we leverage the ergodic metric, which drives agents to spend time in regions proportional to the expected amount of information there. We define the multi-objective sparse sensing ergodic (MO-SS-E) metric in order to optimize when and where each sensor measurement should be taken while planning trajectories that balance the multiple objectives. We observe that our approach maintains coverage performance as the number of samples taken considerably degrades. Further empirical results on different multi-agent problem setups demonstrate the applicability of our approach for both homogeneous and heterogeneous multi-agent teams.
BibTeX
@inproceedings{rao2023multi,
  author = {Ananya Rao and Howie Choset},
  title = {{Multi-Objective Sparse Sensing with Ergodic Optimization}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2023},
  doi = {10.1109/iros55552.2023.10341535},
}
Figure